A methodology to ensure and improve accuracy of Ki67 labelling index estimation by automated digital image analysis in breast cancer tissue

Table 1 Nuclear algorithm settings for the DIA calibration after the Genie classifier

Algorithm setting	DIA-0	DIA-1	DIA-2
Averaging radius (μ)	1	1	1
Curvature threshold	2.5	2.5	2.5
Segmentation type	Cytoplasm rejection	Cytoplasm rejection	Cytoplasm rejection
Threshold type	Edge threshold	Edge threshold	Edge threshold
Lower intensity threshold	0	0	0
Upper intensity threshold	220	230	230
Min. nuclear size (μ²)	20	45	40
Max. nuclear size (μ²)	1,000,000	1,000	1,000
Min, roundness	0.1	0.1	0.1
Min. compactness	0	0	0.2
Min. elongation	0.1	0.1	0.2
Remove light objects	removes no nuclei	removes no nuclei	removes no nuclei
Weak (1+) threshold	210	210	229
Moderate (2+) threshold	188	188	188
Strong (3+) threshold	162	162	162
Black threshold	0	0	0
Edge trimming	Weighted	Weighted	Weighted

DIA-0 (default), DIA-1 (subjective) and DIA-2 (based on quantified bias). Modified settings are highlighted in bold.

ISSN: 1465-542X